training_config.json

{
  "model": {
    "name": "DavidCollective",
    "version": "1.0.0",
    "architecture": "geometric-multi-scale",
    "active_blocks": [
      "down_0",
      "down_1",
      "down_2",
      "down_3",
      "mid",
      "up_0",
      "up_1",
      "up_2",
      "up_3"
    ],
    "num_timestep_bins": 100,
    "num_patterns_per_timestep": 10
  },
  "training": {
    "optimizer": "AdamW",
    "learning_rate": 0.0001,
    "weight_decay": 0.01,
    "scheduler": "CosineAnnealingLR",
    "batch_size": 32,
    "epochs": 20,
    "gradient_clip": 1.0
  },
  "loss": {
    "feature_similarity_weight": 0.5,
    "rose_weight": 0.3,
    "ce_weight": 0.2,
    "pattern_diversity_weight": 0.05,
    "use_soft_assignment": true,
    "temperature": 0.1
  },
  "data": {
    "source": "symbolic-synthesis",
    "num_samples": 10000,
    "complexity_distribution": {
      "1": 0.05,
      "2": 0.15,
      "3": 0.4,
      "4": 0.25,
      "5": 0.15
    }
  }
}